Human Activity Recognition

Human activity recognition (HAR) is essential for healthcare, surveillance, smart homes, and human-computer interaction, but traditional approaches relying on handcrafted features and shallow machine learning models often fail to capture the intricate dynamics of human movements. This study introduces HARNet, a deep learning (DL) model that overcomes these limitations by integrating advanced video preprocessing techniques and state-of-the-art architectures such as CvLSTM, LRCN, and BiLSTM. HARNet employs comprehensive video preprocessing and was evaluated on the UCF50 and HMDB51 datasets, which cover a wide range of human activities. We systematically analyzed the CvLSTM, LRCN, and BiLSTM models to assess their ability to learn spatial and temporal features from video sequences. Experimental results showed that BiLSTM achieved the highest accuracy, with 98.67% on UCF50 and 94.23% on HMDB51, outperforming LRCN and CvLSTM across all metrics, including precision, recall, F1 score, and ROC-AUC, and demonstrating the lowest error rates. HARNet effectively recognizes diverse human activities, enhancing intelligent systems that interpret human behavior with high precision; BiLSTM's bidirectional nature proved especially effective in capturing temporal dependencies, making it ideal for real-world applications in monitoring, security, and human interaction, significantly boosting the performance and reliability of HAR systems.

Members of my family who understood my busy days. In particular, my wife and my parents. The Professors Ricardo Rabelo, Antonio Loureiro, Bruce Thomas and Mark Billinghurst for having patience with me and give me directions to conclude this work. The colleagues who spent some precious time and weekends contributing to the development. In particular, Thiago D'Angelo. To my friends who made my adaptation in Australia easier. There are too many people to cite here... All persons who imposed barriers and make this work harder. I became stronger learning with you. There are too many people to cite here...

Video surveillance is an extensively used tool due to the high rate of atypical behavior and many cameras that enable video capture and storage. Unfortunately, most of these cameras are operator dependent for stored content analysis. This limitation necessitates the provision of an automatic behavior identification system. This behavior identification can be achieved using unsupervised (generative) computer vision methods. Deep learning makes it possible to model human behavior regardless of where they could be. We attempt to classify current research work to report the ongoing trends in human behavior recognition using deep learning algorithms. This paper reviews various aspects, like the ones associated with machine learning and deep learning models, human activity recognition (HAR), deep learning frameworks/tools, abnormal behavior datasets, and a variety of other current trends in the field of automatic learning. All these are to give the researcher a sense of direction in this area.

Cette communication présente une méthode de recherche approximative des plus proches voisins (APPV) modérément sûre mais très efficace. Nous partons d'une méthode de recherche APPV se basant sur des distances entre données quantifiées calculées à l'avance : les PQ-codes. Se plaçant dans le modèle « honnête mais curieux » dans lequel le serveur et l'utilisateur suivent le protocole mais sont capables de déduire de l'information sur les données transitées, nous analysons les menaces en terme de fuite d'information pour cette méthode. Pour empêcher une reconstruction du signal requête par le serveur, nous proposons une version des PQ-codes utilisant deux quantificateurs distincts pour le serveur et l'utilisateur. Les avantages de notre méthode sont (a) une perte significative d'information de la requête par le serveur (b) une base de taille fixée (c) aucune perte dans la qualité de la recherche.

Group abnormal behaviors often occur abruptly under video surveillance, thus bringing serious consequences. How to recognize these behaviors correctly has always been the difficulty in research on intelligence video surveillance. This paper is based on the basic theory of Markov Random Fields to extract the features of those in video images, so as to recognize the group abnormal behaviors under video surveillance. Experiments show that this method can well reflect the real situation at the spot.

Gait analysis has been recognized as an efficient method to help realize human activity recognition; however, there is currently no existing review study focused on wearable activity recognition methods that employ gait analysis in the recognition process. In this study, different wearable-gait-analysis-based (WGA-based) activity recognition methods were summarized and compared from the aspects of wearable sensor types, data segmentation, feature extraction, and classification methods. The limitations of the current research and potential opportunities for future research in this field are also discussed.

Je tiens à remercier sincèrement mes directeurs de recherche, le professeur Shengrui Wang, et la professeure Hélène Pigot, pour la pertinence de leur encadrement, leurs valeureux conseils et encouragements, et surtout pour leurs qualités humaines. Leur rigueur scientifique, leurs conseils judicieux et leur disponibilité tout au long de cette thèse ont joué un rôle déterminant et m'ont permis de mener à terme ce travail de recherche avec les qualités escomptées. Avec mes directeurs de recherche, j'ai appris la flexibilité, l'efficacité et la patience. Ils demeurent des professeurs envers qui j'ai beaucoup de respect. Je remercie également le Conseil de Recherches en Sciences naturelles et en Génie du Canada pour le soutien financier durant deux ans. Je ne peux oublier de remercier ma famille : ma mère, mon père, mes frères et ma soeur pour leur soutien et leurs encouragements durant toute cette période. Mes vifs remerciements vont particulièrement à ma femme. Merci de m'avoir soutenu et encouragé dans les moments difficiles. Merci d'être aussi patiente avec moi. Merci pour tout. ii Abréviations LDA Latent Dirichlet Allocation HMM Hidden Markov Model BN Réseau Bayésien (Bayesian Network) NB Bayésien Naif (Naive Bayes

This review paper examines the integration of Explainable AI (XAI) techniques into abnormal human activity detection from surveillance videos, emphasizing their significance in enhancing transparency, accountability, and trustworthiness in AI attention mechanisms, we discuss how these techniques provide insights into the decision enabling stakeholders to understand and interpret XAI for debugging errors, identifying biases, and increasing trust among human operators in various security contexts. Additionally, we discuss ongoing research directions in XAI for abno development of advanced explanation techniques and the incorporation of user feedback into the explanation process. By fostering collaboration between humans and AI, XAI holds the potential to enhance the effect deployment of surveillance systems, contributing to safer and more secure environments for all stakeholders.

ien que de nos jours, il existe de nombreux outils logiciels conviviaux pour les calculs scientifiques et la simulation de réseaux électriques, on ne disposait pas encore d'outils satisfaisants aux besoins spécifiques de la modélisation et de l'analyse de la qualité de l'énergie électrique. Un tel outil devrait être d'utilisation simple et produire des formes d'ondes analogues dans le temps à celles enregistrées dans le monde réel. Etat actuel de la simulation de la qualité de l'énergie Il existe actuellement un grand nombre de logiciels conviviaux permettant d'étudier le comportement stationnaire de réseaux électriques en ce qui concerne la qualité de l'énergie électrique. Alors qu'ils permettent par exemple d'étudier l'amplitude de chutes de tension, les modèles pour régime permanent ne se prêtent pas à la détermination de la durée de ces chutes. La même limitation existe lorsque les utilisateurs désirent inclure le comportement d'éléments complexes, tels que des fours à arc, des entraînements électriques, des parafoudres, etc., à l'aide du modèle du système. Une autre déficience de ces outils pour régime stationnaire est qu'ils ne permettent pas d'appliquer, de vérifier et de tester des algorithmes de protection et de surveillance fournis par des systèmes fondés sur microprocesseurs. Le programme de transitoires électromagnétiques EMTP (ElectroMagnetic Transients Program) est utilisé couramment par les ingénieurs électriciens pour étudier le comportement transitoire de réseaux électriques. Toutefois, un problème majeur du programme EMTP Le PQ-SMART™ est un outil de planification pour réseaux d'énergie électrique qui combine des produits logiciels usuels du marché en vue de leur application par des utilisateurs disposants de différents niveaux de connaissances techniques. L'intention à la base du PQ-SMART, l'abréviation de Power Quality Simulation, Monitoring And Relay Testing (Simulation de la qualité de l'énergie électrique, surveillance et essai de relais), résidait la réutilisation de logiciels déjà existants et d'usage très courant dans le Groupe ABB. Cela explique le choix de la plate-forme fondée sur PC, avec la technique à fenêtres et les programmes MATLAB et EMTP. Cet outil est destiné à l'analyse et à la modélisation de problèmes de qualité de l'énergie dans des réseaux électriques.

Hidden Markov Models (HMMs) are learning methods for pattern recognition. The probabilistic HMMs have been one of the most used techniques based on the Bayesian model. First-order probabilistic HMMs were adapted to the theory of belief functions such that Bayesian probabilities were replaced with mass functions. In this paper, we present a second-order Hidden Markov Model using belief functions. Previous works in belief HMMs have been focused on the first-order HMMs. We extend them to the second-order model.

Human Activity Recognition (HAR) from sensors has real-world applications such as fall detection of elderly or improvement in human-robot interactions, but privacy is of concern in videos of such applications. We investigated the feasibility of developing a real-time HAR mobile application that detects 3 different human poses: standing, sitting and lying, while preserving privacy by a facial detection method that also masks faces in the video. We proposed a hybrid 1-Dimensional Convolutional Neural Network with Bidirectional Long Short-Term Memory (1DCNN-BiLSTM) algorithm to recognise these 3 poses. We trained our proposed 1DCNN-BiLSTM algorithm using RGB videos from 3 public datasets (URFall, NTU RGB+D, Fall Detection) whereby features were extracted using MoveNet. The results yielded 82.1% test accuracy and 0.930 macro average AUC-ROC score on the combined dataset, hence demonstrating feasibility of deploying the proposed algorithm for balanced as well as imbalanced datasets. We then implemented our proposed 1DCNN-BiLSTM algorithm with face detection and masking in a mobile application. The results showed that the mobile application could perform human pose detection inference in real time (3 people at 23.0-25.3 FPS). Hence this study showed promise of a larger scale-study of these HAR systems, such as monitoring potential falls in video surveillance.

New mobile applications need to estimate user activities by using sensor data provided by smart wearable devices and deliver context-aware solutions to users living in smart environments. We propose a novel hybrid data fusion method to estimate three types of daily user activities (being in a meeting, walking, and driving with a motorized vehicle) using the accelerometer and gyroscope data acquired from a smart watch using a mobile phone. The approach is based on the matrix time series method for feature fusion, and the modified Better-than-the-Best Fusion (BB-Fus) method with a stochastic gradient descent algorithm for construction of optimal decision trees for classification. For the estimation of user activities, we adopted a statistical pattern recognition approach and used the k-Nearest Neighbor (kNN) and Support Vector Machine (SVM) classifiers. We acquired and used our own dataset of 354 min of data from 20 subjects for this study. We report a classification performance of 98...

Automatically recognizing the e-learning activities is an important task for improving the online learning process. Probabilistic graphical models such as Hidden Markov Models and Conditional Random Fields have been successfully used in order to identify a web user activity. For such models, the sequences of observation are crucial for training and inference processes. Despite the efficiency of these probabilistic graphical models in segmenting and labeling stochastic sequences, their performance is adversely affected by the imperfect quality of data used for the construction of sequences of observation. In this paper, a formalism of the possibilistic theory will be used in order to propose a new approach for observation sequences preparation. The eminent contribution of our approach is to evaluate the effect of possibilistic reasoning -during the generation of observation sequences-on the effectiveness of Hidden Markov Models and Conditional Random Fields models. Using a dataset containing 51 real manipulations related to three types of learners' tasks, the preliminary experiments demonstrate that the sequences of observation obtained based on possibilistic reasoning significantly improve the performance of Hidden Marvov Models and Conditional Random Fields models in the automatic recognition of the e-learning activities.

Gait analysis has been recognized as an efficient method to help realize human activity recognition; however, there is currently no existing review study focused on wearable activity recognition methods that employ gait analysis in the recognition process. In this study, different wearable-gait-analysis-based (WGA-based) activity recognition methods were summarized and compared from the aspects of wearable sensor types, data segmentation, feature extraction, and classification methods. The limitations of the current research and potential opportunities for future research in this field are also discussed.

The healthcare industry is facing a number of challenges including skyrocketing costs, medical error incidence, inadequate staffing in hospitals and lack of coverage in rural and underserved urban areas. Healthcare workers are under increasing pressure to provide better services to more people using limited financial and human resources. The one proposed solution to the current crisis is ubiquitous (pervasive) healthcare. The wide scale deployment of wireless networks, wearable computing will improve communication among patients, physicians, and other healthcare workers as well as enable the delivery of accurate medical information anytime anywhere, thereby reducing errors and improving access. At the same time, advances in wireless technologies such as intelligent mobile devices and wearable networks—have made possible a wide range of efficient and powerful medical applications. Ubiquitous healthcare is the application in the healthcare industry by mixing the applications of ubiqui...

Parkinson disease affect bodily functions and there is a growing need for advanced solutions to offer therapeutic advice to patients. A framework using arti-facial intelligence and machine learning techniques has been proposed to address this. The proposed system employs a combination of RGB, inertial, and depth sensors data. The inertial signals have been filtered using a notch filter to obtain the optimal wearable sensor data by examining the upper and lower cutoff frequencies. Multiple features have been calculated, including mel frequency cepstral coefficients (MFCC), statistical features and Gaussian mixture model (GMM) features. On the other hand, silhouettes have been extracted from RGB and depth images, and four crucial parameters have been employed to gauge the level of accuracy with which patients with neurological disorders performed their activities, including the angle formation between the hands and lower/upper half of the body and center of the body, and the angle between the hands of the silhouette. The resulting features have been fused and classified using principle component analysis and a reweighted genetic algorithm. Evaluation using cross-validation on mRI (multi-modal 3d human pose estimation dataset using mmwave, RGB-D, and inertial sensors) and MHEALTH (mobile health) datasets showed a recognition accuracy rate of 97.29% and a 97.94%. The study highlights the need for more datasets to address challenges in rehabilitation using human activity recognition with multi-modal sensors.

Human-human interaction recognition is crucial in computer vision fields like surveillance, human-computer interaction, and social robotics. It enhances systems' ability to interpret and respond to human behavior precisely. This research focuses on recognizing human interaction behaviors using a static image, which is challenging due to the complexity of diverse actions. The overall purpose of this study is to develop a robust and accurate system for human interaction recognition. This research presents a novel image-based human interaction recognition method using a Hidden Markov Model (HMM). The technique employs hue, saturation, and intensity (HSI) color transformation to enhance colors in video frames, making them more vibrant and visually appealing, especially in low-contrast or washed-out scenes. Gaussian filters reduce noise and smooth imperfections followed by silhouette extraction using a statistical method. Feature extraction uses the features from Accelerated Segment Test (FAST), Oriented FAST, and Rotated BRIEF (ORB) techniques. The application of Quadratic Discriminant Analysis (QDA) for feature fusion and discrimination enables high-dimensional data to be effectively analyzed, thus further enhancing the classification process. It ensures that the final features loaded into the HMM classifier accurately represent the relevant human activities. The impressive accuracy rates of 93% and 94.6% achieved in the BIT-Interaction and UT-Interaction datasets respectively, highlight the success and reliability of the proposed technique. The proposed approach addresses challenges in various domains by focusing on frame improvement, silhouette and feature extraction, feature fusion, and HMM classification. This enhances data quality, accuracy, adaptability, reliability, and reduction of errors.

In the fields of body-worn sensors and computer vision, current research is being done to track and detect falls and activities of daily living using the automatic recognition of human actions. In the area of human-machine communication, different combinations of sensors and communication technologies are often used to capture human action. Many researchers have also worked with artificial intelligent systems to detect actions, understand scenes, and implement systems that are more efficient in human action recognition. Although effective approaches are needed to detect outdoor activities with the combination of human actions, feature extraction can be quite a complicated task in a human activity recognition system development. Thus, this paper proposed a solution to detect human activities via hybrid descriptors based on robust features and accurate results. In this study, complex backgrounds, including multiple humans in video frames, were detected. First, inertial signal and video frames are pre-processed using denoising techniques, after which the frames are used to remove the background by detecting human motions and extracting the silhouettes. Then, these silhouettes are further used to extract the human body key points to make the human skeleton. Then the time and frequency domain features are extracted for inertial signals, and geometric features are extracted for the skeleton body points. Finally, multiple feature sets are combined and fed into a zero order optimization model, after which logistic regression is utilized to recognize each action. The proposed system has been evaluated on three benchmark datasets, including, the UP Fall dataset, the University of Rzeszow Fall dataset, and the SisFall dataset and proved its significance by achieving accuracy of 91.51%, 92.98%, and 90.23%, on the aforementioned datasets respectively. INDEX TERMS Fall detection, geometric characteristics, human activity recognition, inertial sensors. In recent years, the demand for improvements in Human Action Recognition (HAR) systems, which aim to tackle The associate editor coordinating the review of this manuscript and approving it for publication was Mohamed Elhoseny . human health and fitness conditions, has increased [1], [2], [3], [4], especially for ambient assisted-living. To increase the performance of their recognition systems, many researchers in this field have utilized inertial sensors, while others have employed depth or RGB sensors [5], [6], [7], [8]. The wide range of applications of HAR systems include

Recognition of human locomotor activities is crucial for monitoring the motion patterns. Current studies for human locomotor activities recognition focused on detecting basic motion patterns. In this study, we proposed a four-modules-based human locomotor recognition model via deep learning, which will support in identifying two signal patterns including static and kinematic motions and classifying the daily activities across different subjects. These motion patterns have been monitored through visual devices along with physical and ambient sensors to extract the complex and basic motion from distinct data forms. The four modules include processing, extraction, optimization, and recognition. Each module focuses on certain processing elements for human locomotion recognition. The processing module represents the preprocessing and segmentation stages for motion and ambient-based data along with extraction of human skeleton points from the visual data. Next, the extraction phase focuses on motion patterns identification and features extraction from the multisensors-based data. Then, the optimization module helps in prominent features selection via genetic algorithm. Furthermore, the recognition module utilized a deep learning technique called hidden Markov model to detect human locomotor activities. The average accuracy rates of 73.05% and 71.14% have been achieved for high-level and atomic-level activities over both datasets. The experimental results have shown that the proposed model outperforms the conventional multisensory systems based on deep classifiers via confidence levels for each skeleton point extracted.

The sensory acquisition of the environment is the most important task of mobile robotics, as it is the foundation for any ability that the robot shall have, later on. Sophisticated tasks require an environment model for path planning, obstacle avoidance and many more. Furthermore, the robot needs to know where it is located within its environment to build-up, complement and update the model. Thus, environment perception, localization and mapping belong to the most important tasks of mobile robotics. We present a general approach supporting the use of arbitrary numbers and types of sensors simultaneously. This allows to operate with a large variety of already existing systems without changing any hardware setup. Furthermore, the semantic environment model, generated by our solution, can directly be used for sophisticated and automated environmental analyses.

Human activity encompasses a series of complex spatiotemporal processes that are difficult to model, but represents an essential component of human exposure assessment. A significant empirical data source like the American Time Use Survey (ATUS) can be leveraged to model human activity, but tractable models require a better stratification of activity data to inform about different, but classifiable groups of individuals that exhibit similar activities and mobility patterns. We have developed a simple unsupervised classification and sequence generation method from existing machine learning algorithms that is capable of generating coherent and stochastic sequences of activity from the data in the ATUS. This classification, when combined with any spatiotemporal exposure profile, allows the development of stochastic models of exposure patterns for groups of individuals exhibiting similar activity behaviors.

Parkinson's disease is a common neurological disease, entailing a multitude of motor deficiency symptoms. In this project, we developed a device with an uploaded edge machine learning algorithm that can detect the onset of freezing of gait symptoms in a Parkinson's patient. The algorithm achieved an accuracy of 83.7% in a validation using data from ten patients. The model was deployed in a microcontroller Arduino Nano 33 BLE Sense Board model and validated in real-time operation with data streamed to the microcontroller from a computer.

Los sensores RGB-D han permitido atacar de forma novedosa muchos de los problemas clásicos en visión por computador, tales como la segmentación, la representación de escenas, la interacción humano-computador, entre otros. Con respecto a la caracterización de movimiento, las estrategias típicas en RGB-D están limitadas al análisis dinámico de formas globales y a la captura de flujos de escena. Estas estrategias, sin embargo, solo recuperan información dinámica entre cuadros consecutivos, limitando el análisis de largos desplazamientos. Este trabajo presenta una estrategia para el cálculo de trayectorias (3D+t), las cuales son fundamentales para la descripción cinemática local, permitiendo una descripción densa de movimiento. Cada trayectoria permite modelar palabras cinemáticas, las cuales en conjunto, describen gestos complejos en los videos. Estas palabras cinemáticas fueron procesadas dentro de un esquema de bolsa-de-palabras para obtener un descriptor basado ocurrencias. Este descriptor de trayectorias logró una exactitud del 80% en 5 gestos y 100 videos.

Unintentional falls are a major public health concern for many communities, especially with aging populations. There are various approaches used to classify human activities for fall detection. Related studies have employed wearable, non-invasive sensors, video cameras and depth sensor-based approaches to develop such monitoring systems. The proposed approach in this study uses a depth sensor and employs a unique procedure which identifies the fall risk levels to adapt the algorithm for different people with their physical strength to withstand falls. The inclusion of the fall risk level identification, further enhanced and improved the accuracy of the fall detection. The experimental results showed promising performance in adapting the algorithm for people with different fall risk levels for fall detection.

Automatic hand gesture recognition plays a fundamental role in current research with the aim of empowering a natural communication between users and virtual reality systems. Starting from an existing work, based on the extraction of two different descriptors from the depth maps followed by their classification with a stand-alone multi SVM classifier, in this paper we improve the gesture recognition system performances and reliability and we evaluate different classification approaches. To this purpose, we first compare the performance of different descriptors and analyze their correlation for assessing their complementarity, and then we demonstrate the advantage gained by their fusion by the Wilcoxon Signed-Rank test. In particular, the novelties of this paper are a new method for extracting features from the curvature image and the design of a very effective ensemble of classifiers to solve the problem.

Human activity recognition (HAR) plays an important role in every spheres of life as it assists in fitness tracking, health monitoring, elderly care, user authentication and management of smart homes. The assistive applications can be implemented on smartphones and wearable watches which are easily accessible and affordable. Now-a-days use of smart phones is ubiquitous, sensor data of diverse physical activities can be easily collected by in-built motion sensors. Many research works are proposed in this area using machine learning techniques including deep neural networks to develop smartphone based applications for human activity recognition. Our objective is to find an effective method from a variety of machine learning including deep learning models for low cost as well as high accuracy activity recognition. To fulfill the objective, a comparative performance study has been done in this work by simulation experiments on five publicly available bench mark data sets. The simulation results show that deep learning models, especially, 2D CNN and BI GRU can be promising candidates for developing smartphone based applications using motion sensor data.

In this paper we address the task of recognizing assembly actions as a structure (e.g. a piece of furniture or a toy block tower) is built up from a set of primitive objects. Recognizing the full range of assembly actions requires perception at a level of spatial detail that has not been attempted in the action recognition literature to date. We extend the fine-grained activity recognition setting to address the task of assembly action recognition in its full generality by unifying assembly actions and kinematic structures within a single framework. We use this framework to develop a general method for recognizing assembly actions from observation sequences, along with observation features that take advantage of a spatial assembly's special structure. Finally, we evaluate our method empirically on two application-driven data sources: (1) An IKEA furniture-assembly dataset, and (2) A blockbuilding dataset. On the first, our system recognizes assembly actions with an average framewise accuracy of 70% and an average normalized edit distance of 10%. On the second, which requires fine-grained geometric reasoning to distinguish between assemblies, our system attains an average normalized edit distance of 23%-a relative improvement of 69% over prior work.

In this paper we address the task of recognizing assembly actions as a structure (e.g. a piece of furniture or a toy block tower) is built up from a set of primitive objects. Recognizing the full range of assembly actions requires perception at a level of spatial detail that has not been attempted in the action recognition literature to date. We extend the fine-grained activity recognition setting to address the task of assembly action recognition in its full generality by unifying assembly actions and kinematic structures within a single framework. We use this framework to develop a general method for recognizing assembly actions from observation sequences, along with observation features that take advantage of a spatial assembly's special structure. Finally, we evaluate our method empirically on two application-driven data sources: (1) An IKEA furniture-assembly dataset, and (2) A blockbuilding dataset. On the first, our system recognizes assembly actions with an average framewise accuracy of 70% and an average normalized edit distance of 10%. On the second, which requires fine-grained geometric reasoning to distinguish between assemblies, our system attains an average normalized edit distance of 23%-a relative improvement of 69% over prior work.

Hearing about the violent conditioning that do on a diurnal base around the world is relatively inviting. particular safety and social stability are seriously hovered by the violent conditioning. A variety of styles have been tried to check the violent conditioning which includes installing of surveillance systems. It'll be of great significance if the surveillance systems can automatically descry violent conditioning and give warning or alert signals. The whole system can be enforced with a sequence of procedures. originally, the system has to identify the presence of mortal beings in a videotape frame. also, the frames which are prognosticated to contain violent conditioning has to be uprooted. The in-applicable frames are to be dropped at this stage. Eventually, the trained model detects violence and these frames are independently saved as images. These images are enhanced to descry faces of people involved in the exertion, if possible. The enhanced images along with other necessary details similar as time and position is transferred as an alert to the concerned authority. The proposed system is a deep literacy grounded automatic discovery approach that uses Convolutional Neural Network to descry violence present in a videotape. But, the disadvantage of using just CNN is that, it requires a lot of time for calculation and is less accurate. Hence, a pre-trained model, MobileNetV2, which provides advanced delicacy and acts as a starting point for the structure of the entire model. An alert communication is given to the concerned authorities using telegram operation.

Hearing about the violent conditioning that do on a diurnal base around the world is relatively inviting. particular safety and social stability are seriously hovered by the violent conditioning. A variety of styles have been tried to check the violent conditioning which includes installing of surveillance systems. It'll be of great significance if the surveillance systems can automatically descry violent conditioning and give warning or alert signals. The whole system can be enforced with a sequence of procedures. originally, the system has to identify the presence of mortal beings in a videotape frame. also, the frames which are prognosticated to contain violent conditioning has to be uprooted. The in-applicable frames are to be dropped at this stage. Eventually, the trained model detects violence and these frames are independently saved as images. These images are enhanced to descry faces of people involved in the exertion, if possible. The enhanced images along with other necessary details similar as time and position is transferred as an alert to the concerned authority. The proposed system is a deep literacy grounded automatic discovery approach that uses Convolutional Neural Network to descry violence present in a videotape. But, the disadvantage of using just CNN is that, it requires a lot of time for calculation and is less accurate. Hence, a pre-trained model, MobileNetV2, which provides advanced delicacy and acts as a starting point for the structure of the entire model. An alert communication is given to the concerned authorities using telegram operation.

Hearing about the violent conditioning that do on a diurnal base around the world is relatively inviting. particular safety and social stability are seriously hovered by the violent conditioning. A variety of styles have been tried to check the violent conditioning which includes installing of surveillance systems. It'll be of great significance if the surveillance systems can automatically descry violent conditioning and give warning or alert signals. The whole system can be enforced with a sequence of procedures. originally, the system has to identify the presence of mortal beings in a videotape frame. also, the frames which are prognosticated to contain violent conditioning has to be uprooted. The in-applicable frames are to be dropped at this stage. Eventually, the trained model detects violence and these frames are independently saved as images. These images are enhanced to descry faces of people involved in the exertion, if possible. The enhanced images along with other necessary details similar as time and position is transferred as an alert to the concerned authority. The proposed system is a deep literacy grounded automatic discovery approach that uses Convolutional Neural Network to descry violence present in a videotape. But, the disadvantage of using just CNN is that, it requires a lot of time for calculation and is less accurate. Hence, a pre-trained model, MobileNetV2, which provides advanced delicacy and acts as a starting point for the structure of the entire model. An alert communication is given to the concerned authorities using telegram operation.

The fast-paced growth of Human Resource Management requires intelligent and sustainable shortlisting models to maximize recruitment efficiency, fairness, and transparency in decision-making. This work proposes a Temporal Convolutional Network based Shortlisting Model to maximize candidate selection and yet remain sustainable, scalable, and ethical recruitment. Contrary to standard machine learning methods, TCN uses dilated causal convolutions to tap into sequential relations among candidate profiles to ensure improved quality and reliability in recruitment decisions. The system is tested using the Kaggle Hiring dataset with attributes including experience, skill, test scores, and previous work history. Hyperparameter search method is applied to find the optimal model to achieve sufficient generalizability among candidate sets. Experimental results portray the TCNbased model as superior to traditional deep learning methods with 99.3% accuracy, 98.9% precision, 98.5% recall, and 98.7% F1-score for candidate selection. The model supports green HRM practice through prejudice reduction, improved efficiency in staffing, provision of diversity, and assistance in long-term staffing planning. The research finds TCN as a powerful method of intelligent shortlisting because it is scalable, equitable, and transparent in recruitment processes. Real-time candidate assessment, adaptive learning aspects, and integration with enterprise HR systems for high-scale deployment will be added in the second generation of work.

6.4 Accuracy (%) when averaging the test-link predictions of models trained on all links from the training room . . . . . . . . . 134 6.5 Accuracy (%) when models are trained with the best or all the links from the other rooms, and tested on the average (median) link or by averaging all test-link predictions . . . . . . . . . . 136 6.6 Median accuracy (%) when R-TTWD is trained with single links from the training room and tested on single links from the test room .

6.4 Accuracy (%) when averaging the test-link predictions of models trained on all links from the training room . . . . . . . . . 134 6.5 Accuracy (%) when models are trained with the best or all the links from the other rooms, and tested on the average (median) link or by averaging all test-link predictions . . . . . . . . . . 136 6.6 Median accuracy (%) when R-TTWD is trained with single links from the training room and tested on single links from the test room .

In this paper we present the implementation of an activity recognition system based on a data acquisition system with sensors for acceleration and physiological parameters. In order to recognize more complex activities we need data acquisition from multiple sensors. Based on our previous experience gained in designing acquisition systems for activity recognition, we have designed and developed a system with three sensors. We developed two possible configurations: one using three acceleration sensors, and one with two acceleration sensors and one heart rate sensor. In the experimental phase we found that the third accelerometer that we've placed on the waist does not provide relevant data to improve the rate of recognition significantly. Thus we opted for the second option, considering that the integration of a physiological sensor provides additional information regarding activities, health status and detecting abnormal states. Starting from the experimental phase, we designed and developed a module integrating the three sensors, with additional functionalities such as an SD card to store the acquired data, and a Bluetooth module for connect to portable devices. In addition, we have integrated a RTC (Real Time Clock) module in order to add time label.

In this paper we present the implementation of an activity recognition system based on a data acquisition system with sensors for acceleration and physiological parameters. In order to recognize more complex activities we need data acquisition from multiple sensors. Based on our previous experience gained in designing acquisition systems for activity recognition, we have designed and developed a system with three sensors. We developed two possible configurations: one using three acceleration sensors, and one with two acceleration sensors and one heart rate sensor. In the experimental phase we found that the third accelerometer that we've placed on the waist does not provide relevant data to improve the rate of recognition significantly. Thus we opted for the second option, considering that the integration of a physiological sensor provides additional information regarding activities, health status and detecting abnormal states. Starting from the experimental phase, we designed a...

This paper presents low cost Inertial Navigation Module, using MEMS sensor and an simple onboard microcontroller data processing (Kalman filters, basic interpretation for data). GPS navigation solutions are investigated, but they encounter big issues in indoor environment. The basic problem is the weak signal power received from satellite in indoor location and therefore the increased error for position. Using another technique and sensor add more precision. MEMS based techniques are promising ones. Using an accelerometer and a digital compass allows the determination of additional parameters that can lead to determine the position or improve existing location. The components used are the usual, available at any specialty store and precision obtained using a minimal data processing software is encouraging. Research direction to follow is to process the data obtained for growth and greater accuracy I.

This paper presents a study of using telepresence and assistance robots in caring for elderly people. We have presented the requirements for these types of robots and possible application fields. Few models of robots will be presented, analyzing technical requirements of assistance robots and their applications. Hereinafter is presented a robot model being under design and development.

The digital data knowledge discovery and data mining due to their immense growth have engrossed a great deal of deliberation in recent years. Numerous applications, such as market investigation and business society, can be promoted by use of the information and facts extracted from a bulky amount of data. Text mining is the skill that allows users to mine useful information from a large amount of digital text documents on the Web or databases.Text Mining is the discovery of unknown information, by automatically extracting information from different written resources. This paper deals with the patterns discovery in text documents using the hidden Markov model. In most existing pattern mining in text mining methods, all go through the problems of lack of accuracy and lack of performance. In the proposed system the first progression is pre-processing step which removes the “noise” word. The next development is the HMMs, which is used for pattern extraction and classification of input d...

One of the imperative mining used in data mining is the association rule mining which mines many eventual information and association from outsized databases. The Association rule mining has many research challenges. The generation of accurate frequent and infrequent itemset by reducing the candidate itemset and reducing the space in the memory is one of the challenges. The generation of positive and negative association rule with high confidence and high quality is another challenge. The next research challenge is to mine an optimized positive and negative association rule. Numerous existing algorithms have been used to overcome these challenges but many such algorithms cause data loss, lack of efficiency and accuracy which also results in redundant rules and loss of space in memory. The major issue in using this analytic optimizing method is specifying the activist initialization limit were the quality of the association rule relays on. In the Proposed work, it is proved that an e...

Ubiquitous Life Care (u-Life care) nowadays becomes more attractive to computer science researchers due to a demand on a high quality and low cost of care services at anytime and anywhere. Many works exploit sensor networks to monitor patient's health status, movements, and real-time daily life activities to provide care services to them. Context information with real-time daily life activities can help in better services, service suggestions, and change in system behavior for better healthcare. Our proposed Secured Wireless Sensor Networkintegrated Cloud Computing for ubiquitous -Life Care (SC 3 ) monitors human health as well as activities. In this paper we focus on Human Activity Recognition Engine (HARE) framework architecture, backbone of SC 3 and discussed it in detail. Camera-based and sensor-based activity recognition engines are discussed in detail along with the manipulation of recognized activities using Context-aware Activity Manipulation Engine (CAME) and Intelligent Life Style Provider (i-LiSP). Preliminary results of CAME showed robust and accurate response to medical emergencies. We have deployed five different activity recognition engines on Cloud to identify different set of activities of Alzheimer's disease patients. I.

Using the UCI-HAR dataset, this paper examines human activity recognition (HAR) from the perspectives of data science and artificial intelligence. The primary objective is to present and evaluate the effectiveness of a multi-layer perceptron (MLP) model, concentrating on six different activity categories. We train and assess the MLP model using the UCI-HAR dataset, contrasting its results with those of convolutional neural networks (CNN). The MLP model shows competitive results, attaining an amazing 97% validation and testing accuracy, highlighting its efficiency for smaller datasets. An extensive study is carried out to assess the model's adaptation to a larger Motion Sense dataset using confusion matrices and cross-entropy, the model shows robustness with an accuracy of 89%. The MLP model performs admirably, demonstrating its capacity to pick up complex patterns. Results from comparative analysis with CNN are competitive, especially when dealing with smaller datasets. The suggested MLP model shows up as a practical and efficient way to advance HAR techniques. Its remarkable performance and versatility not only show its usefulness in real-world scenarios but also point to interesting directions for further study in the area of HAR.

Les chiffres désaisonnalisés sont la source d'information la plus utilisée par les utilisateurs de statistiques officielles, plus particulièrement par les analystes et les chercheurs. Eurostat fournit des séries économiques désaisonnalisées relatives à l'Union Européenne et à la Zone Euro. Comme il est difficile de faire converger les politiques en matière de désaisonnalisation des États membres et d'Eurostat, il est nécessaire de fournir aux utilisateurs des métadonnées sur le processus de désaisonnalisation. Eurostat a récemment commencé de présenter des rapports internes sur la qualité du processus de désaisonnalisation. Cette communication met en évidence la structure et les mesures de la qualité sélectionnées. Quatre types de mesure de la qualité peuvent être envisagées : a) les mesures de la qualité utilisée à des fins de recherche et d'évaluation comme, par exemple, pour les comparaisons de logiciels; b) les mesures de la qualité pour les praticiens de la désaisonnalisation qui ont besoin d'ensembles d'indicateurs pour ajuster leur désaisonnalisation et décider comment elle devrait être établie pour une série individuelle; c) les mesures de la qualité pour la vérification du processus de désaisonnalisation et la production, les lignes directrices etc.; d) les mesures de la qualité pour les utilisateurs finaux qui étudient les données publiées et ne s'intéressent pas aux diagnostics.

This paper introduces a new 3D skeleton-based gait recognition method for motion captured by a low-cost consumer level camera, namely the Kinect. We propose a new representation of human gait signature based on the spatio-temporal changes in relative angles among different skeletal joints with respect to a reference point. A sequence of joint relative angles (JRA) between two skeletal joints, computed over a complete gait cycle, comprises an intuitive representation of the relative motion patterns of the involved joints. JRA sequences originated from different joint pairs are then evaluated to find the most relevant JRAs for gait description. We also introduce a new dynamic time warping (DTW)-based kernel that takes the collection of the most relevant JRA sequences from the train and test samples and computes a dissimilarity measure. The use of DTW in the proposed kernel makes it robust in respect to variable walking speed and thus eliminates the need of resampling to obtain equal-l...

Dans ce travail de thèse, nous nous intéressons à la qualité des informations récoltées par des capteurs sur le web. Ces données forment des séries de données temporelles qui sont incomplètes et imprécises, et sont sur des échelles quantitatives peu comparables. Dans ce contexte, nous nous intéressons plus particulièrement à la variabilité et la stabilité de ces séries temporelles. Nous proposons deux approches pour les quantifier. La première se base sur une représentation à l'aide des quantiles, la seconde est une approche floue. A l'aide de ces indicateurs, nous proposons un outil de visualisation interactive dédié à l'analyse de la qualité des récoltes effectuées par les capteurs. Ce travail s'inscrit dans une collaboration CIFRE avec la société Kantar.This thesis focuses on the quality of the information collected by sensors on the web. These data form time series that are incomplete, imprecise, and are on quantitative scales that are not very comparable. In thi...

Robots for the elderly are a particular category of home assistive robots, helping people in the execution of daily life tasks to extend their independent life. Such robots should be able to determine the level of independence of the user and track its evolution over time, to adapt the assistance to the person capabilities and needs. We present an heterogeneous information management framework, allowing for the description of a wide variety of human activities in terms of multi-modal environmental and wearable sensing data and providing accurate knowledge about the user activity to any assistive robot.

Using the UCI-HAR dataset, this paper examines human activity recognition (HAR) from the perspectives of data science and artificial intelligence. The primary objective is to present and evaluate the effectiveness of a multi-layer perceptron (MLP) model, concentrating on six different activity categories. We train and assess the MLP model using the UCI-HAR dataset, contrasting its results with those of convolutional neural networks (CNN). The MLP model shows competitive results, attaining an amazing 97% validation and testing accuracy, highlighting its efficiency for smaller datasets. An extensive study is carried out to assess the model's adaptation to a larger Motion Sense dataset using confusion matrices and cross-entropy, the model shows robustness with an accuracy of 89%. The MLP model performs admirably, demonstrating its capacity to pick up complex patterns. Results from comparative analysis with CNN are competitive, especially when dealing with smaller datasets. The suggested MLP model shows up as a practical and efficient way to advance HAR techniques. Its remarkable performance and versatility not only show its usefulness in real-world scenarios but also point to interesting directions for further study in the area of HAR.

This paper presents a human action recognition method using dynamic time warping and voting algorithms on 3D human skeletal models. In this method human actions, which are the combinations of multiple body part movements, are described by feature matrices concerning both spatial and temporal domains. The feature matrices are created based on the spatial selection of relative angles between body parts in time series. Then, action recognition is done by applying a classifier which is the combination of dynamic time warping (DTW) and a voting algorithm to the feature matrices. Experimental results show that the performance of our action recognition method obtains high recognition accuracy at reliable computation speed and can be applied in real time human action recognition systems.